Artificial intelligence has been sold for years as a productivity engine for offices, decision-making and the economy as a whole. According to Forrester analyst JP Gownder, speaking to The Register in mid-January, the evidence so far does not support that claim: AI has not yet delivered its central promise of measurable productivity gains.
95 percent of pilots without return
Gownder points to a 2025 MIT study that found roughly 95 percent of enterprise generative AI pilots produced no meaningful financial return or measurable impact on profit and loss. Large companies have spent heavily on AI tools and projects, but in most cases the results have not shown up in their numbers.
The pattern also appears at the national level. Despite widespread adoption of AI tools across industries, recent US productivity statistics show no clear signal that can be attributed to the technology.
The Solow paradox returns
Economists have seen this before. In 1987 the Nobel laureate Robert Solow observed that the computer age was visible everywhere except in the productivity statistics. Heavy investment in information technology through the 1970s and 1980s took years to register in national output, and the eventual gains came only after businesses reorganized how they worked around the new tools.
Gownder and other analysts argue AI is following a similar path. The technology is available and widely deployed, but organizations have not yet changed their processes, structures and skills enough to turn it into output.
Weak returns, real job losses
The uncomfortable part of Forrester’s analysis is that weak productivity results have not slowed workforce changes. Forrester forecasts that AI and automation will eliminate about 6 percent of US jobs by 2030, roughly 10.4 million positions. Gownder describes these losses as structural and permanent rather than cyclical: the roles will not return in their current form, even if the technology that replaced them has not yet made companies measurably more productive.
Forrester frames the overall effect as real but modest, and says AI will mostly augment workers rather than replace them. Even so, the combination of cuts without clear gains raises a question of who benefits during the transition.
Implementation, not technology
The analysts’ conclusion is that the obstacle is not the capability of the models but how they are deployed. Without deeper organizational change, including redesigned workflows, retraining and clear measurement of outcomes, AI risks remaining an expensive experiment for many companies. The history of the Solow paradox suggests the gains may eventually come, but it also suggests they will be earned through management decisions rather than delivered automatically by the software.
